Assigned is not adopted — the Copilot number nobody checks

Open the Microsoft 365 admin centre and you can establish, in about four clicks, how many Copilot licences you bought and how many you handed out. Both numbers are correct. Neither answers the question you will be asked at renewal.

That question is some version of is this working? — and assignment cannot answer it, because assignment is something IT does, not something a user does.

Three numbers, not one

A Copilot estate has three populations, and the distance between them is where the money sits.

• Purchased — what finance is billed for every month.
• Assigned — what has actually been attached to a named person.
• Active — what somebody used, in a real app, inside a window you chose.

Two gaps follow, and they are not the same problem.

The gap between purchased and assigned is procurement debt: seats bought for a phase two that slipped, for leavers nobody replaced, for a department that changed its mind. Nobody is using them because nobody has them. It is embarrassing, and it is easy — a reconciliation and one conversation with whoever signs the renewal.

The gap between assigned and active is the hard one. Those seats are attached to real people who are not using them. Closing it means a decision about a person rather than a line item, which is precisely why it tends not to get made.

Why assignment looks like success

Rollout programmes get measured by assignment because assignment is the part IT controls. You can plan it, schedule it, report it green and close the project. "300 of 300 licences deployed" is a sentence you can put in a steering pack.

Usage is nobody's deliverable in the same way. It depends on training that happened once, on whether a manager uses it in front of their team, on whether the work someone does is the kind Copilot helps with at all. So it gets watched for a month after go-live and then quietly stops being watched.

The bill does not stop.

Decide what "active" means before you measure it

Active is not a fact you discover, it is a definition you commit to. Two choices make it, and both are yours.

The signal. Opening the Copilot app is not the same as using Copilot in Word. Microsoft exposes per-app activity in its Microsoft 365 Copilot usage report, and the per-app view is the one worth keeping — a user who is "active" only because the Copilot chat page loaded on their phone is not evidence of anything.

The window. Thirty days flatters a tenant on a holiday month and punishes seasonal roles. Ninety is defensible but slow to react. Pick one, write down why, and keep it the same between scans — a definition that moves makes every trend meaningless.

The two gaps have different owners

This is the practical reason to separate them rather than quote one "utilisation" percentage.

• Purchased but unassigned is a licensing and procurement job. The fix is to release the seats or assign them, and it needs nobody's permission beyond the budget holder's.
• Assigned but dormant is an enablement job first and a reclaim job second. Some of those users need twenty minutes with someone who does the same role and already uses it. Some of them genuinely do not need the licence. You cannot tell which from the licence table alone.

Report them as one number and you will get one answer — usually "run more training" — applied to a population where half of it will not help.

The number that survives the renewal meeting

Finance does not want an adoption percentage. It wants a figure in money per month, with a date on it, and some evidence that the figure moved last time somebody acted on it.

That means two things in practice. Express the dormant population as spend, not as a count — a reclaimable monthly figure converts directly into the renewal conversation. And keep the previous scan, so the current one is a comparison rather than another standalone snapshot. A single number invites an argument about methodology. A trend invites a decision.

Where this leaves the admin centre

The admin centre is not wrong, it is just answering a different question. It is a system of record for entitlement. Adoption is a question about behaviour, and behaviour has to be measured against a definition you chose, over a window you fixed, per app, and then compared to the last time you looked.

Copilot Insights does that part: purchased, assigned and actively-used seats side by side, adoption broken down by app, a dormancy threshold you set yourself, and the reclaimable spend expressed in money per month and per year. It connects with read-only Microsoft Graph consent and reads activity metadata only — never prompt or file content — so there is nothing in the way of a security sign-off.

Related: how many quiet days make a seat reclaimable, why adoption never stalls everywhere at once, and what the renewal meeting actually needs.